game theory, optimal stopping, probability and statistics || the square-root game
TRANSCRIPT
The Square-Root GameAuthor(s): David BlackwellSource: Lecture Notes-Monograph Series, Vol. 35, Game Theory, Optimal Stopping, Probabilityand Statistics (2000), pp. 35-37Published by: Institute of Mathematical StatisticsStable URL: http://www.jstor.org/stable/4356079 .
Accessed: 15/06/2014 04:09
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp
.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].
.
Institute of Mathematical Statistics is collaborating with JSTOR to digitize, preserve and extend access toLecture Notes-Monograph Series.
http://www.jstor.org
This content downloaded from 91.229.229.203 on Sun, 15 Jun 2014 04:09:23 AMAll use subject to JSTOR Terms and Conditions
The Square-root Game
David Blackwell
University of California
Department of Statistics
367 Evans Hall # 3860
Berkeley, California 94720-3860
U. S. A.
e-mail:[email protected]
Abstract
In this work I give an elementary proof of the following : "The
absolute 1/2 moment of the beta (1/4,1/4) distribution about t is
independent of t for 0 < t < 1.
Keywords : beta distribution, two person game, Richardson extrapo-
lation.
Theorem 1 The absolute 1/2 moment of the beta (1/4,1/4) distribution
about t is independent of t for 0 < t < 1 :
/ p(x)(\x-t\{ll2))dx, Jo
where
p(x) = [x(l-z)](-3/4\
is independent oft for 0 < t < 1.
More generally, for any a with 0 < a < 1, the absolute a ? th moment of the beta ((1
? a)/2, (1
? a)/2) distribution about t is independent of t for
0<t<l.
This note, which I am pleased to write in honor of my old friend Tom
Ferguson, is about the process that led to the Theorem.
Quite a few years ago, shortly after Tom Ferguson got his first PC, I
asked him about the Square-root Game :
35
This content downloaded from 91.229.229.203 on Sun, 15 Jun 2014 04:09:23 AMAll use subject to JSTOR Terms and Conditions
36 David Blackwell
Square-root Game
Players I and II simultaneously choose numbers ? and y in the unit inverval.
Then II pays I the amount \x -
y\^l^2\
Later that day, Tom told me that the value of the game is .59907, to 5
places. I asked him how he got such accuracy, since he could solve games
only up to 30 ? 30 on his machine. He said that he'd used Richardson
extrapolation (which I'd never heard of). He told me a bit about Richardson
extrapolation, and we turned to other rhings.
Then, in my Fall 1994 game theory class I assigned, as a homework prob-
lem, to solve the Square-root Game to 3 places. Several students succeeded,
but one group of four students, working together, claimed to have solved
the game to 15 places. According to them, if either player used the beta
(1/4,1/4) strategy, then Player I's expected income, as calculated by Math-
ematica, was constant to 15 places, no matter what the other Player did.
They could not prove that a beta (1/4,1/4) strategy gave a constant income,
and neither could I.
Later I asked Jim Pitman about the more general case as stated in the
Theorem, and he gave a not-quite-elementary proof. Later he found a gen-
eralization to higher dimensions in Landkof [1972]. Finally I found an ele-
mentary proof, given below.
The method the four students used to get their solution is simple and
instructive.
1. They solved a discrete version, restricting each Player to the 21 choices
0, .05,..., .95,1. The good strategy for each player was a [/-shaped
distribution, symmetric about 1/2.
2. They calculated the variance of this distribution, and found the beta
distribution symmetric about 1/2 with the same variance. It was beta
(.2613, .2613).
3. They guessed that .2613 was trying to be .25, so tried beta(l/4,1/4) as a strategy.
Here is the proof of the Theorem. Fix a, 0 < a < 1, and put
/(*)= ?lv{x){\x-t\a)dx Jo
where p(x) = [x(l -
?)]^a+1^21
This content downloaded from 91.229.229.203 on Sun, 15 Jun 2014 04:09:23 AMAll use subject to JSTOR Terms and Conditions
The Square-root Game 37
We must show that / is constant on 0 < t < 1. Its derivative is
f'(t) = a [\(t -
??a-%(?) dx - a [ [(? -
?)(a_1)]?(?) dx
With the change of variable u = 1 ? ? in the second integral, we get
/'(t) = a[?(? -
?)(a~?)}?(?) dx- f ~\(1 - t -
u^-^IpCu) du]
= a(F(?)-F(l-?)), where
F(t) = /t[(i-x)(a-1)]p(x)dx. Jo
So we must show that F(l ?
t) = F(t). To evaluate F, make the lineax
fractional change of variable ? = (? ?
x)/(t ?
tx) : ? = t(l ?
^)/(l ?
tz) (see
Carr, [1970], Formula 2342). We get
F(t) = [i(l -
t)]?a-W [\(1 -
2)(-(a+1)/2)][^a"1)] dz jo
So F(l -1) = F(i), proving the Theorem.
So the value of the Square-root game is I's expected income when he chooses
? according to beta (1/4,1/4) and II chooses y = 0, namely
G(1/2)/G(1/4)G(1/4)) /1(x(1/2))([x(l-x)](-3/4))dx = G(1/2)G(3/4)G(1/4)
Jo = .599070117367796....
So Tom's first five places were correct.
Aknowledgement
I thank the referee for his kind remarks.
References
[1] CARR, G. S. (1970). Formulas and Theorems of Pure Mathematics.
Chelsea.
[2] LANDKOF, N. S. (1972). Foundations of Modern Potential Theory.
Springer-Verlag. N. S.
This content downloaded from 91.229.229.203 on Sun, 15 Jun 2014 04:09:23 AMAll use subject to JSTOR Terms and Conditions